Assessing the Relationship Between Groundwater Nitrate Concentrations and Environmental Variables Through Repeat Sampling and Statistical Machine Learning: Dutch Flats, Nebraska

Authors

Martin Wells, University of Nebraska-LincolnFollow

First Advisor

Troy E. Gilmore

Date of this Version

9-2018

Document Type

Article

Comments

A THESIS Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Master of Science, Major: Environmental Engineering, Under the Supervision of Professor Troy E. Gilmore. Lincoln, Nebraska: September, 2018

Copyright (c) 2018 Martin J. Wells

Abstract

Nitrate-contaminated aquifers are common in landscapes dominated by agricultural land use. Health concerns related to consuming nitrate-contaminated groundwater are well documented and continued research aimed at decreasing concentrations is critical. A 1990s U.S. Geological Survey (USGS) study focused on groundwater characteristics in the Dutch Flats area of western Nebraska. Agricultural-related practices were determined to largely influence groundwater recharge and nitrate concentrations ([NO₃^-]). Since the conclusion of the USGS study, a transition to more efficient irrigation technology has been observed in this region. The emphasis of this 2016 study was to resample several well nests examined in 1998 to determine whether shifts in water resources management have (1) reduced groundwater recharge rates, (2) increased biogeochemical processes, and (3) reduced groundwater [NO₃^-]. Though 2016 ³H/³He age-dating indicated an increase in groundwater age and decreased recharge rate (19.3 years; 0.35 m/year; n = 8), the mean values were not statistically different from 1998 (15.6 years; R = 0.5 m/year). Samples of δ¹⁵N-NO₃^- and dissolved oxygen (n=14) did not indicate major changes in biogeochemical processes, including denitrification. Long-term [NO₃^-] data from the North Platte Natural Resources District (NPNRD) showed 60% of wells sampled in both 1998 and/or 1999 and 2016 (n = 87) had decreased in [NO₃^-], though median concentrations were not statistically different. Given the extensive long-term NPNRD nitrate dataset (n=1,049), we also applied statistical machine learning to (1) evaluate the method as a means to estimate groundwater lag time, (2) assess the influence of 15 predictor variables on Dutch Flats groundwater [NO₃^-], and (3) evaluate the validity of the model through comparisons with field investigations. Overall, Random Forest displayed promising results for evaluating Dutch Flats groundwater [NO₃^-], though limitations were discovered when modeling temporal data.

Advisor: Troy E. Gilmore